(Packable) Pack and unpack byte buffers faster

[pvdrz]

This PR changes the Packable implementations of Vec<T>, Box<[T]>, VecPrefix<T, B> and BoxedSlicePrefix<T, B> in order to specialize them when T is equal to u8.

Before this changes. packable would do something like

for item in self.iter() {
     item.pack(packer)?;
}

to pack any byte buffer, meaning that each byte will trigger one write in the packer. With the changes proposed here. Byte buffers are packed and unpacked in a single read/write. Some benchmarks show a 6x performance improvement.

I ran the test suite of packable using miri given that this code relies on unsafe heavily, No issues were found.

[thibault-martinez]

Why is a lot of format being changed when the CI wasn't complaining about it ?

[pvdrz]

Why is a lot of format being changed when the CI wasn't complaining about it ?

Because we didn't have a rustfmt.toml before. That's why cargo +nightly fmt was not merging the imports.

[thibault-martinez]

Why is a lot of format being changed when the CI wasn't complaining about it ?

Because we didn't have a rustfmt.toml before. That's why cargo +nightly fmt was not merging the imports.

Ah my bad, I missed the fact you added it.